The following versions of Transformation Portal are currently supported with security updates:

If you discover a security vulnerability in Transformation Portal, please DO NOT open a public issue. Instead:

1. GitHub Security Advisory (Preferred): Create a private security advisory at https://github.com/RC219805/Transformation_Portal/security/advisories/new
2. Direct Contact: Reach out via GitHub (@RC219805)
3. Include:
   • Affected version(s)
   • Steps to reproduce
   • Potential impact assessment
   • Your contact information for follow-up

• Initial Response: Within 48 hours
• Status Update: Within 5 business days
• Resolution Target:
  • Critical: 7 days
  • High: 14 days
  • Medium: 30 days
  • Low: Next release cycle

1. Acknowledgment: You'll receive confirmation that we've received your report
2. Assessment: Our security team will evaluate the vulnerability
3. Communication: We'll keep you informed throughout the resolution process
4. Credit: With your permission, we'll acknowledge your contribution in the fix announcement

This repository uses:

• Dependabot: Automated dependency updates for security vulnerabilities
• Code Scanning: CodeQL analysis on every PR
• Secret Scanning: Prevents accidental credential commits
• Security Advisories: Private vulnerability reporting via GitHub
• Branch Protection: Main branch requires security checks to pass
• Workflow Token Permissions: All workflows use least-privilege permissions: declarations
  • contents: read (default) - Read-only repository access
  • contents: write - Only for dependency submission and automated PR creation
  • security-events: write - CodeQL and security scanning only
  • pull-requests: write - AI code review bot only

Given our image/video processing nature, special attention is required for:

• File Upload Security:

  • Maximum file size limits (default: 500MB for images, 5GB for videos)
  • Strict MIME type validation
  • Magic number verification for file formats
  • Filename sanitization to prevent path traversal

• TIFF Processing:

  • Validation of TIFF tags to prevent buffer overflows
  • Limits on image dimensions (max 65536x65536)
  • Protection against compression bombs

• Depth Anything V2 Model: Validate input dimensions to prevent memory overflow (max 4096x4096)
• Point Cloud Generation: Limit vertex count to prevent DoS (max 10M vertices)
• Temporary File Management: Secure cleanup of intermediate depth maps
• GPU Memory: Monitor and limit VRAM usage (default: 8GB max)

• Model Files:

  • Only load models from trusted sources
  • Verify model checksums before loading
  • Sandboxed model execution environment recommended

• Depth Pipeline:

  • Input size restrictions to prevent OOM attacks
  • Rate limiting for API endpoints
  • Secure temporary file handling for intermediate outputs

• Supply Chain:

  • All dependencies use version constraints to balance security and compatibility
  • For security-critical deployments, consider strict version pinning (e.g., via lock files)
  • Regular dependency audits via pip-audit (governed scanner in CI)
  • Automated security scanning in CI/CD pipeline

• Recent Security Updates:

  March 2026:

  • PyTorch CVE-2025-32434 - Critical RCE vulnerability via torch.load()
    • Supported-lane remediation: macOS Apple Silicon ML core lock rotates to torch==2.8.0 / torchvision==0.23.0
    • Frozen-lane posture: Linux and macOS Intel remain unsupported/frozen historical ML lanes
    • Defense in depth: Runtime enforcement of weights_only=True remains mandatory for all torch.load() calls
    • Implementation: Use transformation_portal.core.security.torch_security.safe_load()
  • Hugging Face Trainer advisory GHSA-69w3-r845-3855
    • Disposition: Managed inference paths do not use transformers.Trainer, Seq2SeqTrainer, TrainingArguments, _load_rng_state, or training-resume flows
    • Action: Dependabot alerts are dismissed as not_used with repo search evidence instead of forcing a transformers 5.x pre-release upgrade into inference stacks
  • Pillow>=10.3.0 - Fixed CVE-2024-28219 (buffer overflow vulnerability)
  • cryptography==46.0.5 - Fixed GHSA subgroup attack vulnerability (SECT curves)
  • black==26.3.1 - Fixed arbitrary file writes from unsanitized cache names

  January 2026:

  • protobuf 6.34.0 - Fixed CVE-2026-0994 / GHSA-7gcm-g887-7qv7 (Dependabot #69)
  • Workflow Hardening - Stricter token permissions across all GitHub Actions workflows
  • Quality Gate - Fixed duplicate permissions block (aa555e0a)

• Security vs Determinism Policy:

  • Transformation Portal prioritizes reproducibility over latest versions (ADR-032)
  • Supported-lane security fixes prefer controlled baseline rotations over opportunistic broad upgrades
  • Version upgrades only occur during controlled baseline rotations
  • All torch.load() calls MUST use weights_only=True parameter

• Known Vulnerabilities (Mitigated):

  • Supported Apple Silicon lane runs on torch 2.8.0 / torchvision 0.23.0
  • Frozen macOS Intel ML lane remains unsupported and gated behind TP_ALLOW_MACOS_INTEL_ML=1
  • requirements/ml-core-linux.txt is a frozen unsupported historical lane and must not drive supported-lane remediation
  • All model loading uses safe_load() wrapper or explicit weights_only=True
  • Pillow: Critical for image parsing vulnerabilities
  • NumPy: Monitor for numerical computation exploits

• Temporary CVE Exceptions (Awaiting Upstream Fix):

  • CVE-2026-4539 (Pygments <=2.19.2): DoS via inefficient regex in AdlLexer
    • Status: No upstream fix available as of March 2026
    • Impact: Low - affects syntax highlighting only (typer → rich → pygments)
    • Exposure: CLI help text rendering, not production image processing
    • Mitigation: CI pip-audit configured with --ignore-vuln CVE-2026-4539
    • Tracking: pygments/pygments#3058
    • Action Required: Remove exception when fixed Pygments is released

If exposing Transformation Portal as a service:

• Authentication: Implement API key or OAuth 2.0
• Rate Limiting:
  • Default: 100 requests/minute per IP
  • Heavy operations: 10 requests/hour
• Input Sanitization: All user inputs must be validated
• Output Filtering: Ensure no metadata leakage in processed files

# If using Flask/FastAPI
headers = {
    "X-Content-Type-Options": "nosniff",
    "X-Frame-Options": "DENY",
    "X-XSS-Protection": "1; mode=block",
    "Strict-Transport-Security": "max-age=31536000; includeSubDomains",
    "Content-Security-Policy": "default-src 'self'; img-src 'self' data:; style-src 'self' 'unsafe-inline'"
}

Security features may impact performance:

• File validation: +100-500ms per upload
• Model checksums: +2-5s on first load
• Input sanitization: +50-200ms per request
• Memory clearing: +10-20% processing overhead
• Depth map bounds checking: +50ms per frame

Note: These overheads are configurable and can be tuned based on your security requirements

# Run with minimal privileges (recommended)
sudo -u nobody python -m transformation_portal.cli

# Or use systemd service with User directive:
# [Service]
# User=nobody
# Group=nogroup

# Use read-only filesystem where possible
docker run --read-only --tmpfs /tmp transformation_portal:latest

# Enable security headers if web-facing
X-Content-Type-Options: nosniff
X-Frame-Options: DENY
Content-Security-Policy: default-src 'self'

Note: The project currently uses config/default_config.yaml for depth pipeline settings (see actual structure with depth_model.variant, processing.zone_tone_mapping, optimization.memory_limit_gb, etc.). The following represents recommended security-related configuration fields that should be implemented for production deployments:

# Recommended security configuration (not currently implemented)
# These settings should be added to application configuration for production use
security:
  max_file_size: 524288000  # 500 MiB (500 * 1024 * 1024 bytes)
  allowed_extensions: ['.jpg', '.png', '.tiff', '.mp4', '.mov']
  temp_directory: '/tmp/transformation_portal'
  cleanup_interval: 3600  # seconds

  depth_processing:
    max_input_dimension: 4096
    max_vertices: 10000000
    memory_limit_gb: 8  # GB (see: optimization.memory_limit_gb in default_config.yaml)

• EXIF Data: Option to strip all metadata from outputs
• Watermarking: Support for invisible watermarks for tracking
• Temporary Files: Secure deletion with multi-pass overwrite
• Memory: Clear sensitive data from memory after processing

Before submitting PRs:

# Run code quality and security checks
make quality-check

# Run full test suite
make test-full

# Install governed security tools from requirements/security.txt
pip install -r requirements/security.txt

# Run static security analysis
bandit -r src/ -ll

# Run dependency vulnerability scan
pip-audit

In case of a security breach:

1. Isolate: Immediately isolate affected systems

   • Disable affected endpoints
   • Revoke compromised credentials

2. Assess: Determine scope and impact

   • Identify affected versions
   • Review access logs
   • Determine data exposure

3. Notify: Alert users within 72 hours if data was compromised

   • GitHub Security Advisory
   • Email to affected users (if applicable)
   • Update security status page

4. Patch: Deploy fixes with priority

   • Emergency patch for critical vulnerabilities
   • Coordinate disclosure with reporters

5. Review: Post-mortem and update security measures

   • Document lessons learned
   • Update security policies
   • Implement additional monitoring

• Python 3.10+ (older versions have known vulnerabilities)
• FFmpeg 6+ (addresses multiple CVEs from earlier versions)
• Operating System with DEP/ASLR support
• Minimum 8GB RAM to prevent swap file exposure
• GPU drivers with security updates (NVIDIA 525+ for CUDA operations)

• HTTPS only for any network operations
• Disable unnecessary network features in production
• Firewall rules to restrict outbound connections
• No telemetry or phone-home features by default

No formal security audits have been conducted yet. This section will be updated as audits are completed.

This project aims to maintain compliance with:

• CWE Top 25: Addressing most dangerous software weaknesses
• OWASP Top 10: Web application security (if applicable)
• PCI DSS: Not applicable (no payment processing features)
• GDPR: For EU user data protection (metadata handling)
• AI Security: Following OWASP ML Security Top 10

Security scanning tools are governed in CI via requirements/security.txt:

# Install governed security tools (bandit, pip-audit)
pip install -r requirements/security.txt

# Run dependency vulnerability scan
pip-audit

# Run static security analysis
bandit -r src/

# Additional optional tools (not governed)
pip install semgrep
semgrep --config=auto

# Existing project tools
pylint --enable=security

# Container scanning (if using Docker)
# Install trivy: https://github.com/aquasecurity/trivy
trivy image transformation_portal:latest

Note: pip-audit and bandit are the governed security tools installed from requirements/security.txt in CI. Additional tools like semgrep can be installed separately as needed for security auditing.

We support responsible disclosure and will:

1. Not pursue legal action against security researchers acting in good faith
2. Work collaboratively to understand and resolve issues
3. Publicly acknowledge researchers (with permission)
4. Maintain a hall of fame for security contributors
5. Consider bug bounties for critical findings (case-by-case basis)

Primary: Create a security advisory at https://github.com/RC219805/Transformation_Portal/security/advisories/new GitHub: @RC219805 Response Time: 48 hours maximum

• docs/guides/BEST_PRACTICES.md - General best practices for contributors
• docs/version_history/changelog.md - Version history and security updates
• docs/architecture/ARCHITECTURE.md - System architecture and security considerations

Last Updated: March 2026 Next Review: June 2026 Security Policy Version: 1.1